Re: [Bug-tar] wishlist: transparent compression on file- and not archive level

[Christian Hammers]

Hello

On 2006-07-19 Doug McLaren wrote:
> On Wed, Jul 12, 2006 at 10:57:58AM +0200, Christian Hammers wrote:
> 
> | For backups, many people decide against the usual compressing method
> | via "-j"/"-c" as one single tape/disk medium error makes the complete
> | tar archive unusable.
> 
> Actually, there's a trivial way around this, and then one that's a bit
> more work.
> 
> Trivial: simply use bzip2 rather than gzip.  bzip2 compresses data in
> block (usually 900 KB in size), and each block is handled
> independantly.  If one block is corrupted, the next block will be
> unaffected, and decompression can restart there.

This would be enough, theoretically. In practice though, the experiment
failed:
 - tar czf test.tar.bz2 /home/ch/doc/pics/    # 1200 files with 160MB
 - hexedit test.tar.bz2  # insert 8 zeroes at about 15%
 - bzip2recover test.tar.bz2 # 192 rec00XXX.tar.bz files created
 - bunzip2 rec*.bz2          
 - mv rec00019.tar.bz bad00019.tar.bz
 - bunzip2 rec*.bz2                    # uncompress the remaining
 - cat rec*.tar > recovered.tar
 - tar xvvf recovered.tar
      ...
     -rw-r--r-- root/ch     1007332 2006-05-29 20:54  \
         home/ch/doc/pics/Freunde/Anna (Lousberg) 2.jpg 
     tar: Skipping to next header 
     tar: Error exit delayed from previous errors

Although only 35 files in 17MB were restored, tar aborts.
As told with "string |grep -c home/ch/doc" there should be
1103 files in 166MB in recovered.tar.

So what am I doing wrong here?



> | To circumvent this drawback tar could use libz to zip the individual
> | file chunks inside the tar archive so that in the case of a bit error
> | on the medium only one file gets corrupted - which would be the case
> | in the present implementation anyway.
> 
> I suggested something similar about 12 years ago, except that in my
> case I was suggesting that if something was to be compressed, it would
> just be renamed in the archive --
It was at least a clever idea...

bye,

-christian-